Serving stack for large multi-modal models (LMMs) with stage disaggregation, request routing, and resource autoscaling. - View it on GitHub